Name		Name	Last commit message	Last commit date
parent directory ..
00-Sandbox.ipynb		00-Sandbox.ipynb
01-GettingStartedSequentialModel.ipynb		01-GettingStartedSequentialModel.ipynb
02-MNIST_MLP.ipynb		02-MNIST_MLP.ipynb
03-MNIST_CNN.ipynb		03-MNIST_CNN.ipynb
README.md		README.md
XOR-numpy-network.ipynb		XOR-numpy-network.ipynb
environment.yml		environment.yml
presentation week 02 - NN architecture.pdf		presentation week 02 - NN architecture.pdf
presentation week02 - NN architecture.tex		presentation week02 - NN architecture.tex

README.md

Week 02

Here's what we have so far:

environment.yml: you can install a custom conda environment (see resources in slack channel) that has all the packages we're using by running (in the terminal!) conda env create --file environment.yml. To activate it (and use those packages) run source activate cwc_ann. This has been tested on at least one computer (James) and worked great! Now you can type source activate cwc_ann when you're working on our projects here, and source deactivate to go back to your other packages when you're done.
XOR-numpy-network.ipynb: from Peter Belhumeur's class (though adapted for python 3) this programs a hidden layer neural network in python (numpy) -- no tensorflow. If you want to understand what this hidden layer is doing, this is worth going through. We won't go through it next Tuesday.
00-Sandbox.ipynb is a space for building your own models. Probably makes sense to do this after working through some of the models below.
01-GettingStartedSequentialModel.ipynb is a very simple notebook where we learn about the Sequential() model in keras
02-MNIST-MLP.ipynb uses a sequential model to build a multi-layer perceptron to classify handwritten digits
03-MNIST-CNN.ipynb uses the same MNIST handwritten digit data but uses a convolutional model -- which preserves spatial patterns in the data -- for classification.